2 * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
4 * SPDX-License-Identifier: GPL-2.0-only
10 #include <common/common.h>
11 #include <common/sessiond-comm/sessiond-comm.h>
12 #include <common/uri.h>
13 #include <common/utils.h>
15 #include <common/compat/endian.h>
18 #include "agent-thread.h"
20 #include "lttng-sessiond.h"
25 struct thread_notifiers
{
26 struct lttng_pipe
*quit_pipe
;
32 enum lttng_domain_type domain
;
35 struct agent_protocol_version
{
36 unsigned int major
, minor
;
39 static int agent_tracing_enabled
= -1;
42 * Note that there is not port here. It's set after this URI is parsed so we
43 * can let the user define a custom one. However, localhost is ALWAYS the
44 * default listening address.
46 static const char *default_reg_uri
=
47 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS
;
50 * Update agent application using the given socket. This is done just after
51 * registration was successful.
53 * This will acquire the various sessions' lock; none must be held by the
55 * The caller must hold the session list lock.
57 static void update_agent_app(const struct agent_app
*app
)
59 struct ltt_session
*session
, *stmp
;
60 struct ltt_session_list
*list
;
61 struct agent
*trigger_agent
;
62 struct lttng_ht_iter iter
;
64 list
= session_get_list();
67 cds_list_for_each_entry_safe(session
, stmp
, &list
->head
, list
) {
68 if (!session_get(session
)) {
72 session_lock(session
);
73 if (session
->ust_session
) {
74 const struct agent
*agt
;
77 agt
= trace_ust_find_agent(session
->ust_session
, app
->domain
);
79 agent_update(agt
, app
);
83 session_unlock(session
);
89 * We are protected against the addition of new events by the session
90 * list lock being held.
92 cds_lfht_for_each_entry(the_trigger_agents_ht_by_domain
->ht
,
93 &iter
.iter
, trigger_agent
, node
.node
) {
94 agent_update(trigger_agent
, app
);
100 * Create and init socket from uri.
102 static struct lttcomm_sock
*init_tcp_socket(void)
105 struct lttng_uri
*uri
= NULL
;
106 struct lttcomm_sock
*sock
= NULL
;
108 bool bind_succeeded
= false;
111 * This should never fail since the URI is hardcoded and the port is set
112 * before this thread is launched.
114 ret
= uri_parse(default_reg_uri
, &uri
);
116 LTTNG_ASSERT(the_config
.agent_tcp_port
.begin
> 0);
117 uri
->port
= the_config
.agent_tcp_port
.begin
;
119 sock
= lttcomm_alloc_sock_from_uri(uri
);
122 ERR("agent allocating TCP socket");
126 ret
= lttcomm_create_sock(sock
);
131 for (port
= the_config
.agent_tcp_port
.begin
;
132 port
<= the_config
.agent_tcp_port
.end
; port
++) {
133 ret
= lttcomm_sock_set_port(sock
, (uint16_t) port
);
135 ERR("Failed to set port %u on socket",
139 DBG3("Trying to bind on port %u", port
);
140 ret
= sock
->ops
->bind(sock
);
142 bind_succeeded
= true;
146 if (errno
== EADDRINUSE
) {
147 DBG("Failed to bind to port %u since it is already in use",
150 PERROR("Failed to bind to port %u", port
);
155 if (!bind_succeeded
) {
156 if (the_config
.agent_tcp_port
.begin
==
157 the_config
.agent_tcp_port
.end
) {
158 WARN("Another process is already using the agent port %i. "
159 "Agent support will be deactivated.",
160 the_config
.agent_tcp_port
.begin
);
163 WARN("All ports in the range [%i, %i] are already in use. "
164 "Agent support will be deactivated.",
165 the_config
.agent_tcp_port
.begin
,
166 the_config
.agent_tcp_port
.end
);
171 ret
= sock
->ops
->listen(sock
, -1);
176 DBG("Listening on TCP port %u and socket %d",
183 lttcomm_destroy_sock(sock
);
189 * Close and destroy the given TCP socket.
191 static void destroy_tcp_socket(struct lttcomm_sock
*sock
)
198 ret
= lttcomm_sock_get_port(sock
, &port
);
200 ERR("Failed to get port of agent TCP socket");
204 DBG3("Destroy TCP socket on port %" PRIu16
,
207 /* This will return gracefully if fd is invalid. */
208 sock
->ops
->close(sock
);
209 lttcomm_destroy_sock(sock
);
212 static const char *domain_type_str(enum lttng_domain_type domain_type
)
214 switch (domain_type
) {
215 case LTTNG_DOMAIN_NONE
:
217 case LTTNG_DOMAIN_KERNEL
:
219 case LTTNG_DOMAIN_UST
:
221 case LTTNG_DOMAIN_JUL
:
223 case LTTNG_DOMAIN_LOG4J
:
225 case LTTNG_DOMAIN_PYTHON
:
232 static bool is_agent_protocol_version_supported(
233 const struct agent_protocol_version
*version
)
235 const bool is_supported
= version
->major
== AGENT_MAJOR_VERSION
&&
236 version
->minor
== AGENT_MINOR_VERSION
;
239 WARN("Refusing agent connection: unsupported protocol version %ui.%ui, expected %i.%i",
240 version
->major
, version
->minor
,
241 AGENT_MAJOR_VERSION
, AGENT_MINOR_VERSION
);
248 * Handle a new agent connection on the registration socket.
250 * Returns 0 on success, or else a negative errno value.
251 * On success, the resulting socket is returned through `agent_app_socket`
252 * and the application's reported id is updated through `agent_app_id`.
254 static int accept_agent_connection(
255 struct lttcomm_sock
*reg_sock
,
256 struct agent_app_id
*agent_app_id
,
257 struct lttcomm_sock
**agent_app_socket
)
260 struct agent_protocol_version agent_version
;
262 struct agent_register_msg msg
;
263 struct lttcomm_sock
*new_sock
;
265 LTTNG_ASSERT(reg_sock
);
267 new_sock
= reg_sock
->ops
->accept(reg_sock
);
273 size
= new_sock
->ops
->recvmsg(new_sock
, &msg
, sizeof(msg
), 0);
274 if (size
< sizeof(msg
)) {
276 PERROR("Failed to register new agent application");
277 } else if (size
!= 0) {
278 ERR("Failed to register new agent application: invalid registration message length: expected length = %zu, message length = %zd",
281 DBG("Failed to register new agent application: connection closed");
284 goto error_close_socket
;
287 agent_version
= (struct agent_protocol_version
) {
288 be32toh(msg
.major_version
),
289 be32toh(msg
.minor_version
),
292 /* Test communication protocol version of the registering agent. */
293 if (!is_agent_protocol_version_supported(&agent_version
)) {
295 goto error_close_socket
;
298 *agent_app_id
= (struct agent_app_id
) {
299 .domain
= (enum lttng_domain_type
) be32toh(msg
.domain
),
300 .pid
= (pid_t
) be32toh(msg
.pid
),
303 DBG2("New registration for agent application: pid = %ld, domain = %s, socket fd = %d",
304 (long) agent_app_id
->pid
,
305 domain_type_str(agent_app_id
->domain
), new_sock
->fd
);
307 *agent_app_socket
= new_sock
;
313 new_sock
->ops
->close(new_sock
);
314 lttcomm_destroy_sock(new_sock
);
319 bool agent_tracing_is_enabled(void)
323 enabled
= uatomic_read(&agent_tracing_enabled
);
324 LTTNG_ASSERT(enabled
!= -1);
329 * Write agent TCP port using the rundir.
331 static int write_agent_port(uint16_t port
)
333 return utils_create_pid_file(
334 (pid_t
) port
, the_config
.agent_port_file_path
.value
);
338 void mark_thread_as_ready(struct thread_notifiers
*notifiers
)
340 DBG("Marking agent management thread as ready");
341 sem_post(¬ifiers
->ready
);
345 void wait_until_thread_is_ready(struct thread_notifiers
*notifiers
)
347 DBG("Waiting for agent management thread to be ready");
348 sem_wait(¬ifiers
->ready
);
349 DBG("Agent management thread is ready");
353 * This thread manage application notify communication.
355 static void *thread_agent_management(void *data
)
358 uint32_t revents
, nb_fd
;
359 struct lttng_poll_event events
;
360 struct lttcomm_sock
*reg_sock
;
361 struct thread_notifiers
*notifiers
= data
;
362 const int quit_pipe_read_fd
= lttng_pipe_get_readfd(
363 notifiers
->quit_pipe
);
365 DBG("Manage agent application registration.");
367 rcu_register_thread();
370 /* Agent initialization call MUST be called before starting the thread. */
371 LTTNG_ASSERT(the_agent_apps_ht_by_sock
);
373 /* Create pollset with size 2, quit pipe and registration socket. */
374 ret
= lttng_poll_create(&events
, 2, LTTNG_CLOEXEC
);
376 goto error_poll_create
;
379 ret
= lttng_poll_add(&events
, quit_pipe_read_fd
,
382 goto error_tcp_socket
;
385 reg_sock
= init_tcp_socket();
389 ret
= lttcomm_sock_get_port(reg_sock
, &port
);
390 LTTNG_ASSERT(ret
== 0);
392 ret
= write_agent_port(port
);
394 ERR("Failed to create agent port file: agent tracing will be unavailable");
395 /* Don't prevent the launch of the sessiond on error. */
396 mark_thread_as_ready(notifiers
);
400 /* Don't prevent the launch of the sessiond on error. */
401 mark_thread_as_ready(notifiers
);
402 goto error_tcp_socket
;
406 * Signal that the agent thread is ready. The command thread
407 * may start to query whether or not agent tracing is enabled.
409 uatomic_set(&agent_tracing_enabled
, 1);
410 mark_thread_as_ready(notifiers
);
412 /* Add TCP socket to the poll set. */
413 ret
= lttng_poll_add(&events
, reg_sock
->fd
,
414 LPOLLIN
| LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
420 DBG3("Manage agent polling");
422 /* Inifinite blocking call, waiting for transmission */
424 ret
= lttng_poll_wait(&events
, -1);
425 DBG3("Manage agent return from poll on %d fds",
426 LTTNG_POLL_GETNB(&events
));
429 * Restart interrupted system call.
431 if (errno
== EINTR
) {
437 DBG3("%d fd ready", nb_fd
);
439 for (i
= 0; i
< nb_fd
; i
++) {
440 /* Fetch once the poll data */
441 revents
= LTTNG_POLL_GETEV(&events
, i
);
442 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
444 /* Thread quit pipe has been closed. Killing thread. */
445 if (pollfd
== quit_pipe_read_fd
) {
449 /* Activity on the registration socket. */
450 if (revents
& LPOLLIN
) {
451 struct agent_app_id new_app_id
;
452 struct agent_app
*new_app
= NULL
;
453 struct lttcomm_sock
*new_app_socket
;
454 int new_app_socket_fd
;
456 LTTNG_ASSERT(pollfd
== reg_sock
->fd
);
458 ret
= accept_agent_connection(
459 reg_sock
, &new_app_id
, &new_app_socket
);
461 /* Errors are already logged. */
466 * new_app_socket's ownership has been
467 * transferred to the new agent app.
469 new_app
= agent_create_app(new_app_id
.pid
,
473 new_app_socket
->ops
->close(
477 new_app_socket_fd
= new_app_socket
->fd
;
478 new_app_socket
= NULL
;
481 * Since this is a command socket (write then
482 * read), only add poll error event to only
485 ret
= lttng_poll_add(&events
, new_app_socket_fd
,
486 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
488 agent_destroy_app(new_app
);
493 * Prevent sessions from being modified while
494 * the agent application's configuration is
500 * Update the newly registered applications's
503 update_agent_app(new_app
);
505 ret
= agent_send_registration_done(new_app
);
507 agent_destroy_app(new_app
);
508 /* Removing from the poll set. */
509 ret
= lttng_poll_del(&events
,
512 session_unlock_list();
518 /* Publish the new agent app. */
519 agent_add_app(new_app
);
521 session_unlock_list();
522 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
523 /* Removing from the poll set */
524 ret
= lttng_poll_del(&events
, pollfd
);
528 agent_destroy_app_by_sock(pollfd
);
530 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
537 /* Whatever happens, try to delete it and exit. */
538 (void) lttng_poll_del(&events
, reg_sock
->fd
);
540 destroy_tcp_socket(reg_sock
);
542 lttng_poll_clean(&events
);
544 uatomic_set(&agent_tracing_enabled
, 0);
545 DBG("Cleaning up and stopping.");
546 rcu_thread_offline();
547 rcu_unregister_thread();
551 static bool shutdown_agent_management_thread(void *data
)
553 struct thread_notifiers
*notifiers
= data
;
554 const int write_fd
= lttng_pipe_get_writefd(notifiers
->quit_pipe
);
556 return notify_thread_pipe(write_fd
) == 1;
559 static void cleanup_agent_management_thread(void *data
)
561 struct thread_notifiers
*notifiers
= data
;
563 lttng_pipe_destroy(notifiers
->quit_pipe
);
564 sem_destroy(¬ifiers
->ready
);
568 bool launch_agent_management_thread(void)
570 struct thread_notifiers
*notifiers
;
571 struct lttng_thread
*thread
;
573 notifiers
= zmalloc(sizeof(*notifiers
));
578 sem_init(¬ifiers
->ready
, 0, 0);
579 notifiers
->quit_pipe
= lttng_pipe_open(FD_CLOEXEC
);
580 if (!notifiers
->quit_pipe
) {
583 thread
= lttng_thread_create("Agent management",
584 thread_agent_management
,
585 shutdown_agent_management_thread
,
586 cleanup_agent_management_thread
,
591 wait_until_thread_is_ready(notifiers
);
592 lttng_thread_put(thread
);
595 cleanup_agent_management_thread(notifiers
);